MatMKL_Pardiso: include the right header file to avoid conflicts with petscblaslapack.h

mkl_set_num_threads cannot be used since it is redefined in mkl.h

Do we really want to call mkl_set_num_threa

MatMKL_Pardiso: include the right header file to avoid conflicts with petscblaslapack.h

mkl_set_num_threads cannot be used since it is redefined in mkl.h

Do we really want to call mkl_set_num_threads at runtime? environment variables could be used as well

show more ...

MatMKL_Pardiso: unused variables

MatMKL_Pardiso: pardiso does not handle unsorted IS for Schur complements since Schur entries are specified by booleans (perm)

Bad merge

Merge remote-tracking branch 'origin/dalcinl/fix-mkl_pardiso' into stefano_zampini/feature-pardiso-schur

Conflicts:
src/mat/impls/aij/seq/mkl_pardiso/mkl_pardiso.c

MatMKL_Pardiso: added control parameter 70 to solve for interior problem

Unlike MUMPS, pardiso does not have built-in support for solving the interior problem, so that we have to to it by hand.

Fix

MatMKL_Pardiso: added control parameter 70 to solve for interior problem

Unlike MUMPS, pardiso does not have built-in support for solving the interior problem, so that we have to to it by hand.

Fix for row-major format for symmetric schur complements

show more ...

MatMUMPS: fix for expansion phase

The previous version of the code was computing the expansion phase even with dirichlet solvers. Now fixed

MatMKL_Pardiso: added support for MAT_FACTOR_CHOLESKY and MatSeqSBAIJ formats.

The conversion routines are now responsible to set the freeaij flag

MatMKL_PARDISO: extended support for CHOLESKY for SeqAIJ

Removed MatCopy_ routine, and replaced with Conversions (just for seqaij for the moment)
Code works using ex192.c

Removed unneeded comments

Fixes for MKL_PARDISO

MatMKL_PARDISO: added schur complement support

Now it works just with LU decomposition.
The class needs some housekeeping and some conversion routines need to introduced to enable properly CHOLESKY

MatMUMPS: set schur complement option only if the IS is not empty

MatHeaderReplace() corrupted the -objects_dump array

Reported-by: Torquil Macdonald Sørensen <torquil@gmail.com>

MatHeaderReplace() and MatHeaderMerge() destroy the second matrix argument therefor

MatHeaderReplace() corrupted the -objects_dump array

Reported-by: Torquil Macdonald Sørensen <torquil@gmail.com>

MatHeaderReplace() and MatHeaderMerge() destroy the second matrix argument therefor pass it as
a pointer to it can be zeroed and not mistakenly reused.

show more ...

MatFactor: unique call for the restore API for Schur complement

MatFactor: forgot to move interface API for MatFactorCreateSchurComplement

MatMUMPS: remove schur_second_step from MUMPS data structure. Use a flag instead

MatFactor: move MatMumps APIs for Schur Complement to MatFactor namespace

MatFactor: start moving Schur complement API from MatMumps to MatFactor namespace

Merge branch 'master' into stefano_zampini/feature-pcbddc-saddlepoint

Conflicts:
src/ksp/pc/impls/bddc/bddcprivate.c

Merge branch 'stefano_zampini/somefixes-matdense-lda'

Merge branch 'pr369/Fande-Kong/fixed-comment-space/master'

MatSeqAIJ: allow MatMatMultAddNumeric_SeqAIJ_SeqDense to be performed on dense matrices with lda != number of rows

forgot to remove call to MatDenseRestoreArray in the previous commit

MatSeqDense: allow MatMatMultNumeric_SeqAIJ_SeqDense to be performed on dense matrices with lda != number of rows

MatMUMPS: free unneeded memory

remove cleanupMumps flag from MatDestroy_Mumps() it should always free the memory it allocated
Bug detected while adding features to libMesh

Reported-by: Matt Landreman <matt.landreman@gmail.com>